中文摘要：

为了充分利用环形对称Gabor变换（CSGT）冗余度小，旋转不变性等优点，提高人脸识别技术在实际应用中的有效性和可行性，提出了一种基于CSGT的人脸特征提取改进算法——CSGT多通道纹理加权算法。该算法首先将人脸图像进行CSGT多尺度分析，然后分块提取纹理统计特征，并将多通道特征自适应加权融合，最后使用PcA加权进行降维得到最具鉴别能力的人脸特征。在ORL、Yale和FERET人脸库上进行实验，与传统算法以及CSGT进行对比，实验结果表明：此算法识别率高，数据量小可行性高，对光照、姿态、表情等干扰具有很好的鲁棒性，且灵活适应于不同的人脸库。

英文摘要：

Abstract. Circularly Symmetrical Gabor Transform （CSGT） is a kind of improved Gabor Transform. In addition to the general properties of the Gabor transform, it also has two advantages of small redundancy and rotation invariance. In order to make full use of the characteristics of CSGT and improve the effectiveness and feasibility of human face recognition technology in practical application, this paper proposed a face feature extraction improved algorithm based on CSGT——Weighted Multi-channel Texture of Circularly Symmetrical Gabor algorithm. Firstly, the face image is transformed to CSGT space and analyzed in multi-scale, then the texture statistical feature is extracted and adaptively weighted to combine the characteristics of multi-channel, finally dimensions are reduced using weighted PCA to get the most discriminative human face features. The experiments simulated ORL, Yale and FERET face databases, and compared the pro- posed algorithm with traditional algorithm and CSGT. The results show the advantages and feasibility of the proposed algorithm.